Machine Learning-Enabled Framework for High-Throughput Screening of MOFs: Application in Radon/Indoor Air Separation

Radon and its progeny may cause severe health hazards, especially for people working in underground spaces. Therefore, in this study, a hybrid artificial intelligence machine learning-enabled framework is proposed for high-throughput screening of metal-organic frameworks (MOFs) as adsorbents for radon separation from indoor air. MOFs from a specific database were initially screened using a pore-limiting diameter filter. Subsequently, random forest classification and grand canonical Monte Carlo simulations were implemented to identify MOFs with a high adsorbent performance score (APS) and high regenerability (R %). Interpretability and trustworthiness were determined by variable importance analysis , and adsorption mechanisms were elucidated by calculating the adsorption sites using Materials Studio. Notably, two MOF candidates were discovered with higher APS values in both the radon/N2 and radon/O2 systems compared with that of ZrSQU which is the best-performing MOF thus far, with R % values exceeding 85%. Furthermore, the proposed framework can be flexibly applied to multiple data sets due to good performance in model transfer. Therefore, the proposed framework has the potential to provide guidelines for the strategic design of MOFs for radon separation.

Automated summary

In this study, a hybrid artificial intelligence machine learning-enabled framework is proposed for high-throughput screening of metal-organic frameworks (MOFs) as adsorbents for radon separation. The framework successfully identified MOFs with high adsorbent performance scores and high regenerability, and two MOF candidates with better performance than the current best-performing MOF were discovered. The proposed framework has the potential to provide guidelines for the strategic design of MOFs for radon separation.